Seven AI image generators that create new images from a photo you upload, for faces, styles, poses, outfits and products, ranked: Mage, ChatGPT, Gemini, Midjourney, Ideogram, Leonardo and Adobe Firefly, with reference limits and prices.
Mage is the best AI image generator with a reference photo upload: it saves a face as a Character from one photo and outfits, poses, places and objects as named References, and it runs the two top-ranked GPT Image 2.5 editing models and Nano Banana 2.1 alongside Mango 3, its Mage-exclusive flagship.
ChatGPT Images 2.5 tops both public image-editing leaderboards and keeps people from your reference photos closer to how they look, on every ChatGPT plan, Free included.
Gemini's Nano Banana 2.1 takes up to 14 reference images and keeps up to four characters consistent.
Midjourney's new Edit Model replaced Character Reference and Omni Reference, and takes up to four references.
Every tool here restricts misusing someone else's photo, most require their consent, and the TAKE IT DOWN Act makes publishing AI intimate images of real people without consent a federal crime.
A reference photo tells the model what to keep. The most common uses are:
A face: keep the same person across new scenes, outfits and poses.
A style: match the colors, lighting or medium of an image you like.
A pose or composition: copy the layout or body position of a photo.
An outfit, product or object: put the same jacket, bottle or prop into new images.
An edit: change part of the photo itself, like the background or a hand.
Tools differ most in how many references they take at once and whether they can save one to reuse later.
We looked at how faithfully each tool keeps a face, product or style from the uploaded photo, how many references it takes at once, whether you can save and reuse them, editing tools, privacy, consent rules and price.
Tool | Best for | Starting price |
|---|---|---|
Mage | Best overall | $10/mo (Basic) |
ChatGPT Images 2.5 | Top-ranked edits | Free (limited), Go $8/mo |
Gemini (Nano Banana 2.1) | Many references at once | Free (limited), Google AI plans |
Midjourney | Restyling a photo artistically | $10/mo (Basic) |
Ideogram | Consistent characters with text | Free weekly credits, Plus $20/mo |
Leonardo.ai | Separate face, style and content references | Free (150 tokens/day), $12/mo |
Adobe Firefly | Commercially safe style and composition | Free daily credits, Standard $9.99/mo |
Mage is a browser-based image and video platform built around saving and reusing references.
What works: You upload one clear portrait to create a Character, then call it by its handle in any prompt, and References save an outfit, pose, location or object the same way, so "@sasha wearing @redcoat in @cafe" combines three uploads in one image. It takes up to 15 references on GPT Image 2.5, 14 on Nano Banana 2.1 and 10 on Mango 3, Mage's exclusive flagship model. Inpaint, Face Swap and Outfit Swap edit your photos directly. Mango 3 and Character creation are unlimited from the $30 Pro plan, and images are private by default.
What doesn't: Stacking many references can muddy the result. Mage doesn't allow a Character built from someone's photo without their consent.
Best for: Creators who want to reuse the same faces, outfits and places across a whole series.
ChatGPT Images 2.5, released September 8, 2026, is OpenAI's image model in ChatGPT and, as GPT Image 2.5 Flare and Sunburst, in its API.
What works: GPT Image 2.5 Sunburst and Flare hold the top two spots on both the Artificial Analysis and LMArena image-editing leaderboards, and OpenAI says 2.5 is better at preserving the people and objects in your reference photos. You upload images to a chat, then describe the result, select an area to change it, or comment on the image. It's on every plan, Free included, and Go costs $8 a month.
What doesn't: OpenAI doesn't publish how many reference images ChatGPT takes, there's no named, reusable reference like a Character, so you re-attach images in each chat, and free use is limited. Its policies restrict using a real person's likeness without consent.
Best for: Quick, accurate edits and new images from a photo, with no setup.
Nano Banana is Google's family of image models, and Nano Banana 2.1, released October 6, 2026, is the newest.
What works: Nano Banana 2.1 takes up to 14 reference images, keeps up to four characters consistent and holds up to 10 objects faithful, which suits product scenes and group shots. It ranks joint third on Artificial Analysis's image-editing leaderboard, and you can edit by describing changes in plain language. The Gemini app has a free tier, accepts several uploads at once and can save your own face as an avatar so you don't re-upload it.
What doesn't: Google's help pages still name Nano Banana 2 as the Gemini app's model, so 2.1's full limits are clearest in the API. Free downloads are capped at 1K, usage limits aren't published as a fixed number, and Google bans using someone's likeness to deceive.
Best for: Combining many people or products from photos into one image.
Midjourney's V8.2, the default since July 24, 2026, added an Edit Model that replaces its old Character Reference, Omni Reference and Retexture tools.
What works: The Edit Model takes up to four reference images and edits by instruction, with inpainting and outpainting in its Editor. Image prompts, style references and moodboards turn a photo into a painted, illustrated or cinematic version with Midjourney's polished look. Plans cost $10, $30, $60 and $120 a month, or 20% less billed yearly.
What doesn't: It isn't built for keeping a real face exact, there's no free plan, and images are public unless you pay $60 a month for Stealth mode. Midjourney takes especially strict action on sexualized images of real people.
Best for: Artists who want a photo reimagined in a striking style.
Ideogram is known for accurate text in images, and its 4.5 model is recommended for consistent characters.
What works: Ideogram 4.5 takes up to five reference images, and its Auto mode takes up to 10, so you can keep a character consistent in posters, thumbnails and ads with readable lettering. Plus costs $20 a month, or $15 billed yearly, and includes unlimited character consistency and private generation.
What doesn't: It ranks below the leaders on the image-editing leaderboards, and free accounts get only weekly credits.
Best for: Designers making branded images of a recurring character.
Leonardo splits reference uploads into separate controls in its Image Guidance panel.
What works: Style Reference copies a look, Content Reference keeps the subject and layout, and Character Reference keeps a face, with pose, depth and edge guides for composition. Free users get 150 tokens a day, and paid plans from $12 a month keep images private.
What doesn't: Character Reference works on its SDXL and Phoenix models rather than every model it offers, and it keeps a likeness rather than swapping a face exactly. Free creations are public.
Best for: Users who want fine control over which part of a photo the model copies.
Adobe Firefly takes uploaded images as composition and style references, and its own models are trained on licensed and public-domain content.
What works: Composition reference copies the layout of a photo, and style reference matches its colors and look, while Firefly Image 5 edits with prompts and layers. Adobe calls its own models commercially safe, outputs carry Content Credentials, and Firefly also runs partner models from Google, OpenAI and others. Plans start at $9.99 a month, with free daily generations.
What doesn't: Adobe's own models keep a face less faithfully than ChatGPT or Nano Banana, and the commercially safe promise doesn't cover the partner models.
Best for: Marketing teams that need reference-based images they can safely use commercially.
Every tool here restricts misusing a real person's face, most require their consent, and most also ban making intimate images of real people at all. In the US, the TAKE IT DOWN Act makes publishing nonconsensual intimate images, including AI-made ones, a federal crime, and since May 19, 2026 the law has required platforms to remove them within 48 hours of a valid request, with the FTC enforcing it. Use your own photos, or get clear permission from the person in them.
Mage, for most people. It saves a face as a Character from one photo, saves outfits, poses, places and objects as References, and runs GPT Image 2.5, Nano Banana 2.1 and its Mage-exclusive Mango 3 in one place, from $10 a month.
GPT Image 2 and 2.5 take up to 15 on Mage and 16 inside Ideogram, and Nano Banana 2.1 takes up to 14. Mango 3 takes 10, Ideogram 4.5 takes five, and Midjourney's Edit Model takes four.
Yes. Upload a photo, then describe the image you want or select an area to change. It works on every ChatGPT plan, Free included, with limits.
Yes, with limits. ChatGPT and Gemini have free tiers, Leonardo gives 150 tokens a day and Adobe Firefly offers free daily generations.
Only with their consent. Every tool here restricts misusing a real person's likeness, and most require permission, and publishing intimate AI images of a real person without consent is a federal crime in the US.
